Climate and Development

Carolina Torreblanca

University of Pennsylvania

Global Development: Intermediate Topics in Politics, Policy, and Data

PSCI 3200 - Spring 2026

Agenda

  1. Climate and Development
  2. Dell, Jones & Olken (2012)
  3. Interpreting Logged Variables

Climate and Development

The Planet Is Warming

Who Caused It

Who Suffers

Why Should Climate Affect Development?

  • Agriculture: crop yields fall above optimal temperature thresholds
  • Labor productivity: heat stress reduces output, especially outdoor and manual work
  • Health: heat illness, expanded disease vectors (malaria, dengue)
  • Conflict: resource scarcity increases stress and violence
  • Infrastructure: extreme weather destroys capital

But Rich Countries Seem Unaffected

  • Most wealthy countries are in temperate climates
  • Many have adapted: air conditioning, irrigated agriculture, crop switching, insurance markets
  • Key question: is the effect of temperature the same everywhere, or does it depend on income?

Dell, Jones & Olken (2012)

The Research Question

Does temperature affect economic growth? Does the effect differ by income?

  • Cross-country pattern: hot countries are poor
  • But that correlation cannot support causal inference
  • Hot countries differ in colonization history, institutions, geography, disease burden…
  • Need a cleaner source of variation

Theory

Hypothesis: Higher temperatures reduce growth in poor countries but not rich ones

  • Proposed channels: agriculture, labor productivity, health, conflict
  • Why poor countries? Less capacity to adapt: rain-fed agriculture, weak health systems, fragile institutions

The Identification Problem

Research Design

Key insight: year-to-year temperature fluctuations are plausibly exogenous

  • A country’s average temperature is confounded with history, geography, institutions
  • But whether this year was hotter than usual is essentially random – weather
  • Identification assumption: temperature deviations from a country’s long-run norm affect growth only through economic channels

Data and Model

125 countries, 1950–2003. Temperature from gridded weather data. GDP growth from Penn World Tables.

\[\Delta \ln y_{it} = \beta_1 T_{it} + \beta_2 P_{it} + \alpha_i + \gamma_t + \epsilon_{it}\]

  • \(T_{it}\): mean temperature in country \(i\), year \(t\), relative to that country’s mean over the full sample (1950–2003)
  • \(\alpha_i\): country fixed effects
  • \(\gamma_t\): year fixed effects

Reading the Table

Col (1): no effect on average across all countries.

Col (2): temperature interacted with a poor country dummy. Large and significant (-1.655***).

Marginal effect in poor countries: β_temp + β_interaction = -1.4pp. Same logic as slopes().

Critical Thinking: The Political Instability Story

  • Dell et al. find more irregular leadership transitions in hotter years in poor countries
  • Does heat cause conflict, or do both reflect deeper fragility?
  • The instability result is smaller and less robust than the GDP result
  • Instability may be a mechanism or a consequence of worse economic outcomes – not a clean separate channel
  • Takeaway: mechanism identification is much harder than reduced-form identification

What the Paper Argues

  • The rich-poor asymmetry suggests climate change will widen global inequality
  • Historical emissions came overwhelmingly from rich countries; damages fall on poor ones
  • These are effects on growth rates, not levels – compounding matters enormously
  • A 1.3pp drag sustained over decades means massive foregone development
  • Adaptation is possible but requires investment poor countries cannot self-finance

Interpreting Logged Variables

Why Log? Reason 1: Skewed Distributions

Why Log? Reason 2: Interpretation Changes

Two common cases:

DV is \(\ln Y\) (log levels): \(\hat\beta \times 100\) = approximate % change in \(Y\)

DV is \(\Delta \ln Y\) (log difference = growth rate): \(\hat\beta\) is directly a percentage point change in the growth rate – no multiplication needed, since the DV is already a percent change

Dell et al. use \(\Delta \ln y_{it}\) as DV – so \(\hat\beta = -0.013\) means 1°C above normal reduces the growth rate by 1.3 percentage points directly.

The Cheat Sheet

\(Y\) in levels \(Y\) logged
\(X\) in levels \(\hat\beta\) units \(\hat\beta \times 100\)% change in \(Y\)
\(X\) logged \(\hat\beta / 100\) units per 1% change in \(X\) \(\hat\beta\)% change in \(Y\) per 1% change in \(X\)